PDA

View Full Version : [ASk] What good topic for my undergraduate thesis/research



manufans0607
01-11-2010, 07:17 AM
Hi all,
Please give me an advice to choose what topic for my thesis/research. OpenCL looks interesting for me so I want to know further about this...
My advisor give me an advice to find how we can use GPU to solve some mathematics computation. So, I told him that we can use OpenCL to solve that problem. Can you give me some mathematics computation? Maybe like "system of linear equations" so I can change the number of variable and I can analyze that...Sorry for my bad english..

Thx B4,,

coleb
01-11-2010, 11:22 AM
A really cool and useful thesis idea: an OpenCL kernel code analyzer that could report how much floating point error is accumulating. OpenCL has 16,32,64,128 bit floating point numbers. Knowing when to use what precision is a black art with major performance implications. Something to analyze how much floating point error is accumulating would be invaluable. Extra credit: have it make suggestions of places where the precision can be down-graded for performance, and should be upgraded for accuracy.

I'm working on a piece of scientific code right now when I compile it to the CPU versus GPU I get different answers. Luckily, my errors are quite gaussian in distribution so I'm not going to worry about it. But I would really like to know where they are coming from.

Take a look at this article: http://www.scientificcomputing.com/arti ... 63009.aspx (http://www.scientificcomputing.com/article-hpc-Numerical-Precision-How-Much-is-Enough-063009.aspx)